In conversation with Yang Yuanqing: In the next year, edge AI will improve performance by three times, "this is a conservative estimate in the short term"

The heat of edge AI has been burning from January's CES to the ongoing MWC, showing no signs of cooling down.

But this should only be the beginning of the edge AI battle; we are still far from "intensified competition" and "starting a price war."

Yang Yuanqing, Chairman and CEO of Lenovo Group, stated in his personal keynote that the combined effect of enhanced computing power and model optimization is expected to achieve a threefold overall performance improvement for edge AI in the next 12 months.

After the speech, Yang Yuanqing candidly admitted, "In the short term, this is still a conservative estimate."

He listed his arguments, citing two factors driving edge AI performance: one is computing power, and the other is model efficiency.

Regarding the former, although Moore's Law has stagnated (some even call it ineffective), when AI chips reignite market vitality, Moore's Law may not stop at all, and could even accelerate.

As for the latter, the most intuitive example is the deep-sea giant DeepSeek, which has carved out a bloody path. With various innovations combined, DeepSeek, which is also a 7B parameter model, is more efficient, more precise, and faster.

(Note: On February 25, Lenovo released the YOGA 2025 series new products, locally deploying DeepSeek-7B, becoming the world's first vendor to deploy such a large model on terminal devices.)

Yang Yuanqing mentioned that he observed other AI-related companies' estimates for edge AI might be "even more aggressive than mine":

Achieving a tenfold growth in about 12-18 months.

Edge AI is on the rise, and Lenovo's first step is to "promote AI PCs"

Regardless of how various parties predict specific figures, one undeniable fact is that edge AI is unparalleled and unstoppable.

Moreover, with the emergence of DeepSeek and all subsequent similar "catfish," the development trend of edge AI can be clearly seen as "accelerating," yet it is difficult to discern the curve of its advancement.

As the rapid development of edge AI becomes foreseeable, how should strategies or corresponding businesses aimed at edge AI be planned?

The leader of Lenovo Group directly stated:

The first step is to promote AI PCs.

The essence of AI PCs lies in their ability to run large models deployed locally. For users, they offer sufficient convenience and security guarantees.

Lenovo is undoubtedly the number one PC manufacturer globally and is also the first proposer and publisher in the AI PC track.

Upon arriving at MWC 2025, Quantum Bit discovered that it may also be the undisputed king of AI PC competition—Lenovo released new AI PC products at both CES and MWC, including a rollable screen AI PC and a foldable screen AI PC This is unique among all exhibitors participating in the two exhibitions.

△ CES rollable screen AI PC (left), MWC foldable screen AI PC (right)

The two products mentioned above are undoubtedly the highlights of AI PCs that attracted countless flashes at the two global exhibitions, but they are just the tip of the iceberg in Lenovo's "push for AI PCs" strategy (the former is set for mass production in June, while the latter is still a concept machine).

Facing the consumer market, Lenovo released its first-generation AI PC 11 months ago, and Yang Yuanqing stated at MWC that a new generation of products will be launched soon.

At that time, everyone will see a significant leap in local artificial intelligence inference capabilities, comparable to OpenAI's o1-mini, but at a significantly reduced cost.

Looking beyond the immediate future, Yang Yuanqing's response is: "In three years, over 80% of PCs will be AI PCs."

Two months ago at CES, Lenovo Group Executive Vice President and President of Intelligent Devices Group (IDG) Luca Rossi clearly stated that a wave of AI PC replacements is coming, with the key time point being 2025.

From a glimpse, Lenovo has its own data as evidence; by the end of 2024, Lenovo's AIPC is expected to occupy 15% of the PC market, ahead of the planned goal of 10% by 2025.

At MWC, Yang Yuanqing continued to discuss that AI PCs are just the first step in Lenovo's layout of edge AI.

Further ahead is "turning more terminals into artificial intelligence terminals."

More terminals include all smart devices (or more), whether it's a phone, a watch, or even a ring with some smart capabilities but limited computing power, smart reading, and data collection.

Edge AI is no longer limited to the currently emerging AI PCs and AI phones.

"Lenovo has developed its own capabilities and experience across various terminals and ecosystems," Yang Yuanqing said. "We have insights into Windows PCs, Android phones, tablets, and even the Chrome ecosystem." This is sufficient to study the rapidly changing AI large models and the ever-evolving terminal forms Then achieve cross-device, cross-platform, and cross-operating system capabilities with different forms, different operating systems, and different ecosystems.

Yang Yuanqing spoke with confidence, Lenovo has this capability.

He added that the core of AI is data, and data is the prerequisite for intelligence. Of course, computing power is also essential.

Today, the best edge AI capabilities are carried on personal computers, and the terminals that generate the most data are everyone's smartphones.

Therefore, Lenovo's first step at this stage is to connect the two across devices, ensure good interaction, unify data management, and allow the data generated by smartphones to be processed by personal computers, enabling them to produce intelligence and achieve the desired intelligent results, ultimately benefiting everyone.

Completing this step is still not the end. This step is merely an intermediate process.

Yang Yuanqing's ultimate idea for the development of edge AI is like this:

No matter how many devices you have, the data generated by each person on their own devices is unified. It is unique to you personally and is your own wealth.

(Besides data), computing power is also uniformly managed. You can imagine using the computing power of your own AI PC to process your own data, using your family's storage and computation to save this data.

Ultimately, achieving the personal AI twin that Lenovo has mentioned multiple times in various occasions, empowering people with AI rather than replacing them.

“Edge intelligence absolutely cannot replace cloud intelligence”

Edge AI is the most dazzling trend at the intersection of AI and hardware, but edge AI cannot solve everything.

Yang Yuanqing gave a decisive expression:

Edge intelligence absolutely cannot replace cloud intelligence.

He elaborated that pursuing AGI based on foundational models requires pre-training on data from the entire network, as well as post-training phases and various engineering optimizations...

This requires a large amount of computing power, which cannot be achieved on the edge. The edge can only perform inference applications.

In other words, the directions of service are different.

In the process of utilizing AI, privatized elements need to have privatized processing methods, but as the saying goes, people nowadays want everything without making choices, and the same applies to AI.

It is necessary to simultaneously meet personalized demands for effectiveness, efficiency, response speed, security/privacy protection, and sustainability.

This aligns with Lenovo's proposal for "hybrid AI" at the 2023 Tech World Conference.

Its core logic is to achieve personalized AI services while protecting user privacy through locally deployed intelligent terminals (such as personal computers and smartphones) and private clouds, combined with public large model capabilities.

It grasps Lenovo's traditional advantages in the hardware field while addressing the dual demands for data security and efficiency in the AI era The dual wheels of hardware and models are racing forward, and hybrid AI has already become a "inevitable trend" stamped by Lenovo.

Looking back, from strategic concepts to the implementation in 2024 (launching AI PCs and intelligent agents), and to the demonstration of hybrid AI at MWC in early 2025, the distant lighthouse has come into view in just a year and a half.

It must be said that Lenovo has thought about edge AI very early; after realizing that edge AI cannot be for everything, the layout and progress have been very rapid.

One More Thing

Lenovo's edge AI targets both the consumer market and the enterprise market.

However, as one of the billions of consumers (and also one of the price-sensitive groups, doge), I found that all AI PCs on the market are priced higher than ordinary PCs.

In response, Yang Yuanqing said:

Deploying and running private large models locally requires high-end computers, which are more expensive to buy, but there will be no operating costs in the future. At most, it will just use some electricity (laughs).

This is also a reason why edge intelligence can be promoted and will become more widespread.

Quantum Bit, original title: "Dialogue with Yang Yuanqing: Edge AI will enhance performance by 3 times in the next year, 'this is a conservative estimate in the short term' | Directly hitting MWC 2025"

Risk Warning and Disclaimer

The market has risks, and investment requires caution. This article does not constitute personal investment advice and does not take into account the specific investment goals, financial situation, or needs of individual users. Users should consider whether any opinions, views, or conclusions in this article are suitable for their specific circumstances. Investment based on this is at their own risk